Downloads · 30 days
5
25% of all-time downloads
srivathsanb14/hw2-nn-automl-stairs
hw2-nn-automl-stairs is a image classification model from srivathsanb14. Use it when you need a label for an image. It is set up for autogluon. The card lists the license as mit.
Classify whether a photo contains stairs or not.
Downloads · 30 days
5
25% of all-time downloads
All-time downloads
20
Public
Repo size
86.3 MB
Likes
0
Public
Click a slice to open those files.
.ckpt44.8 MB · 52%
From the Hugging Face model README
Classify whether a photo contains stairs or not.
DataFrame column image = local RGB file paths. predict() returns 0 (non_stairs) or 1 (stairs).
Dataset images are 224x224 RGB. AutoGluon then applies its shorter-side resize, center crop, and ImageNet normalization.
Source: ArinRoths/StairvsNonStair_Dataset (ArinRoths).
| Split | Rows | Notes |
|---|---|---|
| train | 374 | originals + brightness / rotation / contrast / blur variants |
| validation | 5 | 5 unaugmented originals |
| test | 5 | 5 unaugmented originals |
| Setting | Value |
|---|---|
| Framework | AutoGluon MultiModal 1.6.1 |
| Preset | medium_quality; model.names=["timm_image"] |
| Search | resnet18_lr4e-4, resnet18_lr1e-3, efficientnet_b0_lr4e-4, mobilenetv3_small_lr4e-4 |
| Time / trial | 150s |
| Seed | 24679 |
| Early stopping | optim.top_k=3, greedy soup; best val checkpoint kept |
| Selected trial | resnet18_lr1e-3 (backbone resnet18, lr 0.001) |
| Selected val accuracy | 1.0000 |
| Split | Accuracy | Balanced acc. | Weighted F1 | Macro F1 |
|---|---|---|---|---|
| published val | 1.0000 | — | 1.0000 | — |
| published test | 1.0000 (5/5) | 1.0000 | 1.0000 | 1.0000 |
n_test=5: one miss is 20 points.
Railings, shadows, and floor lines mistaken for treads; rotation that crops the stair run; background shortcuts from 32 original photos.
Author reviewed photos for faces/PII. Do not deploy for building access or surveillance.
MIT for this repo. Attribute ArinRoths for the images.
Google Colab T4, 4 x 150s, seed 24679. Python 3.13.15.
from pathlib import Path
import zipfile
from huggingface_hub import hf_hub_download
from autogluon.multimodal import MultiModalPredictor
archive = hf_hub_download("srivathsanb14/hw2-nn-automl-stairs", "autogluon_image_predictor_dir.zip")
out = Path("stairs_model"); out.mkdir(exist_ok=True)
with zipfile.ZipFile(archive) as zf:
zf.extractall(out)
predictor = MultiModalPredictor.load(str(out))
Cursor helped align this notebook with the instructor image AutoGluon card. I chose the classmate dataset, the TIMM/LR search grid, and reviewed every test photo.